I can’t tell you how many times I’ve drawn a sketch of a cat and thought, gosh, I wish I could turn this crappy doodle into an image of a real cat. OK, I’ve never, ever had that thought before, but now there’s a site that does just that, and it will probably make you wonder how you survived without such a glorious tool in your life until now.
Pushbullet co-founder Christopher Hesse created an image-to-image translation demonstration that generates a corresponding photograph from any image you feed it. In this case, though, Hesse trained his system on roughly 2,000 stock images of cats. When you sketch out a cat on his site here, it’ll produce an image of a cat that best matches your sketch.
However, if your cat sketching skills are as “good” as ours, the results can be hilarious, or mildly terrifying.
Grumpy cats are doable.
It can even handle two cats.
This entertaining tool is based on legitimate research, a computer vision idea called pix2pix, or “Image-to-Image Translation with Conditional Adversarial Networks.” (You can find it on Github here.)
Unfortunately, cats really aren’t the best example of how this technique can be used. Hesse notes: “Some of the pictures look especially creepy, I think because it’s easier to notice when an animal looks wrong, especially around the eyes. The auto-detected edges are not very good and in many cases didn’t detect the cat’s eyes, making it a bit worse for training the image translation model.”
That’s OK, though. On his site, there are three other image translation examples you can fool around with, as well: one for building facades, one that generates a shoe from your line drawing, and one that generates handbags. Clearly, though, the cat-creating version is superior to these other options. I mean, come on.
You can head over to Affinelayer.com to try it out yourself.
H/T @ftrain/Twitter